From Future Learning To Current Action: Long-Term Sequential Infrastructure Planning Under Uncertainty

从未来的学习到当前的行动：不确定性下的长期顺序基础设施规划

• 批准号: 1663479
• 负责人: Matteo Pozzi
• 金额: $ 55万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-09-01 至 2022-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1663479&HistoricalAwards=false
• 关键词: Future; Learning; Current; Action; Long

Stakeholders and owners of assets and infrastructure systems exposed to extreme events have to take decisions related to risk control and mitigation. Long-term infrastructure planning need not be fixed in the present time, as decisions can be periodically revised depending on available knowledge. This may suggest postponing critical decisions until uncertainty has been sufficiently reduced, though delays may actually increase risks in the shorter term. By investigating the relationship between expected future learning (in terms of policy and technology, for example) and current action in optimal decision making under uncertainty, this project will allow for a) adaptive optimizing of long-term management of infrastructure systems, b) exploration of flexible approaches to asset design, and c) assessment of the value of collecting additional information on attendant risks. The outcome of this project will contribute to society's ability to select appropriate, adaptive risk mitigation actions in optimally engaging limited resources.Accordingly, this project will investigate how decisions on infrastructure planning should depend on the expected future available information on various sources of risk and uncertainty. This will include the development of methods for formulating realistic assumptions about learning rates, and for integrating these assumptions into scalable schemes for system-level sequential decision optimization under uncertainty. The project will develop a framework for integrating future expected learning into decision making optimization via probabilistic modeling of the effects of various exogenous conditions on infrastructure use and planning. Sequential infrastructure management will be framed as a Partially Observable Markov Decision Process (POMDP) to create an efficient computational framework able to identify optimal policies.

易受极端事件影响的资产和基础设施系统的利益相关者和所有者必须做出与风险控制和缓解相关的决策。长期基础设施规划不需要在目前就确定下来，因为可以根据现有知识定期修订决定。这可能意味着推迟关键决策，直到不确定性充分减少，尽管拖延实际上可能在短期内增加风险。通过研究在不确定性条件下，预期的未来学习（例如政策和技术方面的学习）与最佳决策中的当前行动之间的关系，本项目将允许a）基础设施系统的长期管理的适应性优化，B）探索资产设计的灵活方法，以及c）评估收集有关伴随风险的额外信息的价值。本项目的成果将有助于提高社会在最佳利用有限资源的过程中选择适当的、适应性强的风险缓解行动的能力。因此，本项目将研究基础设施规划的决策应如何依赖于各种风险和不确定性来源的预期未来可用信息。这将包括制定有关学习率的现实假设的方法的发展，并将这些假设整合到可扩展的计划，在不确定性下的系统级顺序决策优化。该项目将开发一个框架，通过对各种外生条件对基础设施使用和规划的影响进行概率建模，将未来预期的学习融入决策优化。顺序的基础设施管理将被构建为部分可观测马尔可夫决策过程（POMDP），以创建一个能够识别最优策略的高效计算框架。

项目成果

Predicting the Evolution of Controlled Systems Modeled by Finite Markov Processes

预测有限马尔可夫过程建模的受控系统的演化

• DOI: 10.1109/tr.2021.3067595
• 发表时间: 2021
• 期刊: IEEE Transactions on Reliability
• 影响因子: 5.9
• 作者: Li, Shuo;Pozzi, Matteo
• 通讯作者: Pozzi, Matteo

Model-free reinforcement learning with model-based safe exploration: Optimizing adaptive recovery process of infrastructure systems

基于模型的安全探索的无模型强化学习：优化基础设施系统的自适应恢复过程

• DOI: 10.1016/j.strusafe.2019.04.003
• 发表时间: 2019
• 期刊: Structural Safety
• 影响因子: 5.8
• 作者: Memarzadeh, Milad;Pozzi, Matteo
• 通讯作者: Pozzi, Matteo

Culture and cognition: Understanding public perceptions of risk and (in)action

文化和认知：了解公众对风险和（行动）的看法

• DOI: 10.1147/jrd.2019.2952330
• 发表时间: 2020
• 期刊: IBM Journal of Research and Development
• 影响因子: 1.3
• 作者: Allen, T.;Wells, E.;Klima, K.
• 通讯作者: Klima, K.

Creating a water risk index to improve community resilience

创建水风险指数以提高社区复原力

• DOI: 10.1147/jrd.2019.2945301
• 发表时间: 2020
• 期刊: IBM Journal of Research and Development
• 影响因子: 1.3
• 作者: Klima, K.;El Gammal, L.;Kong, W.;Prosdocimi, D.
• 通讯作者: Prosdocimi, D.